AI glossary: terms explained simply

The key terms around AI in the enterprise, each with a short definition, an everyday comparison and an example.

RAG (Retrieval-Augmented Generation)

RAG (retrieval-augmented generation) is a method in which a language model first retrieves relevant passages from a document collection and then bases its answer on those passages. This lets the AI work with a company's own, current knowledge without retraining the model, and every statement can be traced back to a source.

MCP (Model Context Protocol)

MCP (Model Context Protocol) is an open standard that lets AI applications such as Claude connect to external data sources and tools. An MCP server exposes, for example, a document archive or a database, and any compatible AI assistant can use that source without a custom integration being built for every combination.

Hallucination (AI)

A hallucination is a statement by an AI language model that is fluent and convincing but factually wrong or not supported by any source. It happens because language models generate likely sequences of words and do not check facts by themselves. Source-based methods such as RAG reduce the risk but do not eliminate it.

Embedding

An embedding is a representation of a text, image or record as a list of numbers (a vector) that captures its meaning. Content with similar meaning gets similar vectors. This lets a semantic search recognise, for example, "notice period" and "time limit for terminating the contract" as related, even though the words differ.

Hybrid search

Hybrid search is a search method that combines classic keyword search (for example BM25) with semantic vector search based on embeddings, and merges the results of both into one ranked list. It finds exact terms such as case numbers or legal sections as well as passages that express the same idea in different words.

On-premise AI

On-premise AI means running AI models and the related software on servers the company itself controls, in its own data centre or a dedicated environment. Data and queries do not leave the company network. This makes it easier to meet GDPR and confidentiality obligations, but it requires own hardware and ongoing operation.

Relational foundation model

A relational foundation model is a pre-trained AI model that makes predictions directly from several linked database tables, such as customers, contracts and claims. Unlike classic machine learning, it does not need the tables to be laboriously flattened into one, and often needs no separate training for each task. A well-known example is KumoRFM.

Text-to-SQL

Text-to-SQL is a method in which an AI model translates a question in natural language into an SQL query that runs against a database. This lets finance or accounting staff analyse data without knowing SQL. Because the generated query is visible, it is possible to trace how a result was produced.

Cancellation early warning (insurance churn prediction)

Cancellation early warning is an AI-based prediction that uses contract, customer and contact data to estimate which insurance policies are likely to be cancelled within a given period. It gives a probability per policy and the main contributing factors, so that sales teams can reach out in a targeted way before the cancellation arrives.

EU AI Act (Artificial Intelligence Act)

The EU AI Act is Regulation (EU) 2024/1689, which governs the development and use of AI systems in the European Union. It classifies AI systems by risk: certain practices are prohibited, high-risk systems are subject to strict obligations, and others mainly face transparency duties. The regulation applies directly in all member states, including Austria.